Research map / Label-efficient and robust learning / Domain adaptation and generalization
High-Level Semantic Feature Matters Few-Shot Unsupervised Domain Adaptation
Lei Yu, Wanqi Yang, Shengqi Huang, Lei Wang, Ming Yang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Replace local features with task-specific semantic features | IMSE · DomainNet 5-way 1-shot, average | accuracy: 48.86 -> 52.79 | Improved |
| Add cross-domain self-training to task-specific semantic embedding (ablation) | TSE · DomainNet, sketch → clipart, real, quickdraw, painting | accuracy: 61.98 -> 68.25; 60.00 -> 61.15; 52.21 -> 58.31; 51.62 -> 53.34 | Improved |
| Add cross-attention to task-specific semantic embedding (ablation) | TSE · DomainNet, sketch → clipart, real, quickdraw, painting | accuracy: 61.98 -> 62.74; 60.00 -> 60.54; 52.21 -> 53.64; 51.62 -> 54.23 | Improved |
| Add task-specific semantic embedding to cross-domain self-training (ablation) | Cross-domain self-training · DomainNet, sketch → clipart, real, quickdraw, painting | accuracy: 57.07 -> 68.25; 53.31 -> 61.15; 41.93 -> 58.31; 46.66 -> 53.34 | Improved |
| Add cross-attention and cross-domain self-training to TSE (ablation) | TSE · DomainNet, sketch → clipart, real, quickdraw, painting | accuracy: 68.25 -> 69.45; 61.15 -> 65.00; 58.31 -> 62.25; 53.34 -> 56.51 | Improved |
| Add semantic feature alignment loss (ablation) | TSECS · DomainNet, sketch → clipart, real, quickdraw, painting | accuracy: 66.83 -> 68.04; 58.29 -> 63.98; 56.51 -> 59.13; 44.25 -> 55.39 | Improved |
| Add class matching loss (ablation) | TSECS · DomainNet, sketch → clipart, real, quickdraw, painting | accuracy: 66.67 -> 68.04; 58.84 -> 63.98; 56.91 -> 59.13; 43.28 -> 55.39 | Improved |
| Add similarity pattern alignment loss to semantic feature alignment and class matching (ablation) | TSECS · DomainNet, sketch → clipart, real, quickdraw, painting | accuracy: 68.04 -> 69.45; 63.98 -> 65.00; 59.13 -> 62.25; 55.39 -> 56.51 | Improved |
| Add all three domain alignment and class matching losses (ablation) | TSECS · DomainNet, sketch → clipart, real, quickdraw, painting | accuracy: 66.67 -> 69.45; 58.84 -> 65.00; 56.91 -> 62.25; 43.28 -> 56.51 | Improved |
| Use cross-domain self-training without semantic embedding (ablation) | TSECS · DomainNet, sketch → clipart | accuracy: 69.45 -> 57.07 | Hurt |
| Train on DomainNet and evaluate on miniImageNet | TSECS · miniImageNet 5-way 1-shot, real → sketch | accuracy: 48.52±0.81 -> 49.83±0.96 | Improved |
Closest papers by meaning
- Prototypical Cross-Domain Self-Supervised Learning for Few-Shot Unsupervised Domain Adaptation (CVPR 2021 · similarity 0.89)
- Cross-domain Few-shot Classification via Invariant-content Feature Reconstruction (IJCV 2026 · similarity 0.89)
- Adapter Naturally Serves as Decoupler for Cross-Domain Few-Shot Semantic Segmentation (ICML 2025 · similarity 0.88)
- Discriminative Sample-Guided and Parameter-Efficient Feature Space Adaptation for Cross-Domain Few-Shot Learning (CVPR 2024 · similarity 0.87)
- Universal Domain Adaptation for Semantic Segmentation (CVPR 2025 · similarity 0.87)
- Domain-Rectifying Adapter for Cross-Domain Few-Shot Segmentation (CVPR 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
